Large Language Models for Knowledge-Intensive Domains: Methods, Systems, and Evaluation

The rapid emergence of Large Language Models (LLMs) and generative AI is reshaping how people and organizations access, synthesize, and apply knowledge. This impact is especially pronounced in knowledge-intensive domains, meaning settings where users pose complex, high-context questions; where answers must be grounded in large, evolving repositories; where relevance and ranking matter as much as generation; and where trust, provenance, and personalization are essential.Three example domains—among many others—illustrate both the promise and the practical challenges, offering similar opportunities for LLM innovation alongside domain-specific hurdles:• Education (higher education and professional learning): AI tutors, content generators, and assessment assistants are reshaping pedagogy, enabling more adaptive feedback, scalable support, and learner-centered resource discovery. At the same time, educational applications demand stringent standards for factuality, pedagogical appropriateness, transparency, and long-term impact on skill development.• E-commerce search and recommendations: Conversational product discovery, personalized ranking, catalog understanding, and customer service automation are transforming digital commerce. These applications require robust query understanding, reliable grounding in product and policy data, careful balancing of relevance and personalization objectives, and responsible handling of fairness, manipulation risks, and consumer trust.• Across these and other knowledge-intensive settings, recent field research suggests that generative AI can function as a "cybernetic teammate," or an agentic AI, replicating some benefits traditionally associated with human collaboration, such as access to diverse perspectives and cross-functional synthesis. This raises important research questions about how LLMs reshape individual performance, team dynamics, domain expertise boundaries, and organizational workflows in both learning and commercial contexts.This call for contributions invites original research on the design, implementation, and evaluation of LLM-based systems for knowledge-intensive applications across consumer, business, enterprise, and government contexts. We welcome work spanning diverse domains and problem settings, including (but not limited to) education, e-commerce, healthcare, law, finance, enterprise knowledge management, scientific discovery, and public services. We are particularly interested in methods and insights that transfer across domains, including how architectures, evaluation protocols, and human-AI interaction patterns generalize (or fail to generalize) when deployed in real-world, high-stakes knowledge workflows. Core technical challenges are shared across domains. Factual accuracy and hallucination mitigation remain critical. In education, fabricated content undermines learning outcomes. In commerce, inaccurate product information or misleading recommendations damage customer relationships and can introduce compliance risk. Retrieval, relevance, and reranking are likewise central. Educational platforms must surface resources appropriate for a learner's level and goals, while e-commerce systems must balance relevance, personalization, and business objectives in ranking and recommendations. Finally, evaluation remains a major bottleneck. Traditional offline metrics often fail to capture nuanced quality dimensions of LLM-mediated experiences, motivating new frameworks that assess grounding, utility, calibration, robustness, and alignment with user intent.Our goal is to foster a broad, interdisciplinary conversation that advances reliable, effective, and scalable LLM-powered systems in knowledge-intensive environments.We invite original research that addresses core technical challenges in knowledge-intensive domains, encompassing innovative methods, systems, and evaluation approaches. Authors are encouraged to propose additional domains where knowledge quality, grounding, and ranking are central. Topics include, but are not limited to:• Retrieval-Augmented Generation (RAG) for domain-grounded answering and synthesis• Deep Research systems, agentic AI search, and other search, discovery, and knowledge work paradigms• Indexing, representation learning, and scalable retrieval over heterogeneous repositories (text, tables, images, logs)• Search relevance modeling and neural reranking for knowledge-intensive queries• LLM-enhanced recommendation systems (resource recommendation, product discovery, next-best-action)• Query understanding at scale: intent detection, classification, query rewriting, clarification strategies• Knowledge integration: knowledge graphs, ontologies, citation and provenance, structured and unstructured fusion• Domain adaptation: fine-tuning, instruction tuning, continual learning, parameter-efficient methods• Hallucination detection, mitigation, and other model output content quality problems: grounding checks, uncertainty calibration, abstention, verification pipelines• Evaluation frameworks and benchmarks: beyond accuracy, including usefulness, pedagogical value, trust, calibration, robustness• Human–AI interaction and teaming: AI as collaborator, workflow design, oversight, expertise boundaries• Conversational systems: dialogue management, memory and personalization, multi-turn search and recommendation• Responsible and ethical AI: bias and fairness in ranking, privacy, manipulation risks, trust, and disclosure• Deployment and monitoring: drift detection, guardrails, cost and latency tradeoffs, A/B testing, and field evaluations Topic editors: Giuseppe (Pino) Di Fabbrizio, Surya Kallumadi, Andrei Lopatenko Source: https://www.frontiersin.org/research-topics/78064/large-language-models-for-knowledge-intensive-domains-methods-systems-and-evaluation